11 - Operations and Sustainment
PHYSICS-INFORMED NEURAL NETWORKS FOR ULTRASONIC LAMB WAVE RECONSTRUCTION IN NON-DESTRUCTIVE TESTING
L. Jilke¹, M. Schubert¹, N. Hosters², F. Raddatz¹, M. Behr², G. Wende¹; ¹German Aerospace Center (DLR), Germany ;²RWTH Aachen University, Germany
We propose a Physics-Informed sinusoidal representation network (PI-SIREN) embedding elastodynamic wave equations for reconstructing sparse ultrasonic guided wavefield data in the field of non-destructive testing. By capturing high-frequency oscillations, PI-SIREN significantly outperforms data-driven baselines and standard PINNs in ultrasonic data-scarce wavefield reconstruction.
DOI: 10.71945/icas2026_0573